Decagon CTO: Fine-Tuned Smaller Models Outperform Frontier LLMs on Specific Tasks
kimberlywtan · x · 2026-08-01
a16z hosted the co-founders of Decagon, an enterprise AI customer support agent company, to discuss their experience deploying AI agents within major banks, airlines, and telcos.
Key takeaways include:
- Smart-vs-cheap models is a false trade-off: A 'dumber' smaller model, when fine-tuned for a specific task, can actually outperform state-of-the-art frontier models while being cheaper and faster.
- Model migration path: For new use cases, it's best to start with frontier models to validate the workflow, then migrate to open-source models as the task matures.
- Reshaping support supply and demand: AI drastically lowers support costs, allowing enterprises to affordably handle previously unmet massive customer service demand. Currently, 90% of Decagon's operations run on open-source models.
Related event: a16z x Decagon: Open-Source Migration and Moat in Enterprise AI Agents(6 posts)→
More from coding & agent
- Connected steps are not a workflow: nobody designs the handoff — usehive · 2026-10-16
- Making Blender animations with Opus, Astra and Argon: how you prompt still matters — chongdashu · 2026-10-16
- Running PolyStrata + Cline coding agent on a $0-fancy laptop at ~90 tok/s — rpwoerk · 2026-10-16
- How do you attribute agent gains: model training or infrastructure? — jrxna · 2026-10-16
- Feed Claude your books, blogs and top engineers' code for claude.md — 'it now writes better code than me' — blaizedsouza · 2026-10-16
- Open-source /label plugin auto-names and colors Claude Code sessions in one step — hermanschutte · 2026-10-16